The AMD HD4890

Spring means new GPU’s and today we are taking a look at the new HD4890 from AMD. The HD4890 has higher clockspeed than the HD4870 1 GB and we of course have tested them both to see what kind of performance increase we get.

INTRODUCTION

Springtime usually means a new GPU from both NVIDIA and AMD and even though we live in troubled times it is that time again in 2009. A week ago in the last days of March I sat down at an AMD presentation where we were shown among other products the card that we will talk about today, the new HD4890 which is using the “new” RV790 GPU. The planned release were on the 9th of April, even though reports were already coming in that cards using it were out in the Far East, but due to the “Green’s” decision to try to spoil the party with a “limited” launch of the new GTX275 AMD decided to move the release forward a week to today.

In this article we will give you an overview of the GPU as well as give you benchmark scores of it compared to the HD4870 1 GB and the GTX260 216 Core. At the same time we also have a review of the new Asus HD4890 Voltage Tweak Edition as well as an article about NVIDIA’s new GPU, the GTX275 making this a interesting day for everyone looking for a new graphics card.

The RV790/HD4890 is not a competitor to the NVIDIA GTX285 or GTX295. Instead it is a chip that is aimed straight to the enthusiast market with cards selling around $250. It slots in neatly above the HD4870 1 GB while leaving the high-end to the HD4870X2. AMD actually showed us an interesting chart which shows the demand they are seeing at the various price points.

While the demand naturally is for the cheaper cards there still is a nice peak around sub-$300 which is the market they hope to aim the HD4890 for.
 

SPECIFICATIONS

If you expected the RV790 to be a 40nm GPU then you are wrong, it still is a 55nm GPU. In fact, looking at the following chart it shows that there isn’t much difference between the RV770 and the RV790.

There is a slight difference in transistor count but the AMD-representative did not want to go into what changes had been made. Overall it seems that the main advantage of the RV790/HD4890 is a slightly faster GPU. In fact, except for the slightly higher speeds there is not much new with this GPU compared to the older GPU’s.

The features are still the same:

Stream computing
AMD’s answer to NVIDIA’s PhysX. Right now the main focus is to help improve the speed of video-encoding. AMD already have a free utility that lets you convert using the GPU but during our briefing they also showed a new program that they are developing with Cyberlink which will have a much nicer interface. That program will however cost a small sum.

UVD2
With acceleration of 2 HD-video streams, full 7.1 audio over HDMI and improved video quality the HD4800 GPU’s are perfect for anyone who also want to watch video with their card.

DirectX10.1
Finally games are coming out that have DirectX10.1 support. The idea is that DX10.1 allows the developers to achieve the same effect as in DX10 but with less work.

Overclocking

One feature that AMD is a bit ambiguous about is overclocking. On one hand they are claiming that the HD4890 should in most cases be overclockable to 950 MHz or more (a 100+ MHz overclock of the GPU), on the other hand they always, even when providing users with the AMD Overdrive tool, say that overclocking will void your warranty.

It looks though that with the big potential for overclocking we should see a lot of factory overclocked cards coming out.

TESTING & METHODOLOGY

To test the HD4890 we did a fresh install of Vista 32. We loaded all the latest drivers for the components in the computer and made sure the OS was updated.
All our games and applications were installed and the drive was then cloned with Acronis True Image so that we could start from a “fresh” OS for every benchmark.
 

Test Rig

Synthetic Benchmarks & Games

3DMark Vantage

www.futuremark.com/benchmarks/3dmarkvantage/features/

The newest video benchmark from the gang at Futuremark. This utility is still a synthetic benchmark, but one that more closely reflects real world gaming performance. While it is not a perfect replacement for actual game benchmarks, it has its uses. We tested our cards at the ‘Performance’ setting.

Currently, there is a lot of controversy surrounding NVIDIA’s use of a PhysX driver for its 9800 GTX and GTX 200 series cards, thereby putting the ATI brand at a disadvantage. Whereby installing the PhysX driver, 3DMark Vantage uses the GPU to perform PhysX calculations during a CPU test, and this is where things get a bit gray. If you look at the Driver Approval Policy for 3DMark Vantage it states; “Based on the specification and design of the CPU tests, GPU make, type or driver version may not have a significant effect on the results of either of the CPU tests as indicated in Section 7.3 of the 3DMark Vantage specification and white paper.” Did NVIDIA cheat by having the GPU handle the PhysX calculations or are they perfectly within their right since they own Ageia and all their IP? I think this point will quickly become moot once Futuremark releases an update to the test.

No ruspirses here. THe increased clockspeed of the HD4890-GPU pulls it 7-10% ahead of the HD4870 1 GB. It’s not very far ahead of the GTX260 216 Core.

Far Cry 2

Again we see a performance advantage for the HD4890 over the HD4870 1 GB. However, it cannot beat the GTX260 216 Core which obviously means it won’t beat the GTX275.

Stalker Clear Sky

There’s a new stand alone benchmark out for the new title Stalker Clear Sky and even though we had a tight deadline with the HD4890-article we wanted to get some Stalker action in. We ran the Stalker Benchmark at it’s extreme settings with AA turned off in Direct X 10 and DX10.1 (for the AMD-cards) mode.

The first benchmark run is set during the day. At the Extreme setting (and boy does it look pretty even without AA) the AMD-cards perform very well, both easily beating the GTX260. We found out that it didn’t matter if we turned on/off DX10.1 – we got the same scores.

 The next benchmark run is set during the night. The result is the same as before.

 The third benchmark run is set during the night while it is raining and a thunderstorm is raging. The result again is the same as before.

Last but not least we got the same benchmark as before, this time with “SunShafts” – meaning a lot of sunlight. It looks like this setting is more CPU-limited than GPU-limited.

Tom Clancy’s HAWX

Tom Clancy’s HAWX is a arcade-flightsim with gorgeous graphics. The game supports DX9, DX10 and DX10.1 and we choose to test DX10 and DX10.1 (for the AMD-cards). We set everything to max, turned on 8xMSAA and used the in-game benchmark to get the average framerate.

 This is a game that really likes the AMD-cards in DX10-mode. Again we did not notice any difference in performance when we switched DX10.1 on or off. The HD4890 performace >20% better than the HD4870 1 Gb and oblitirates the GTX260 216 Core.

Battleforge

This is yet another new game that supports DirectX10.1. The game is a interesting mix between card-game and RTS and sports really nice visuals. This is also one of the few games we’ve seen where a AMD-logo pops up during the start of the game. We set every setting to max and turned on 8xMSAA.

With abmient occlusion (SSAO) turned off the AMD-cards again crushes the GTX260 216 Core. The difference between the HD4870 1 Gb and the HD4890 is around 5-8%.

With SSAO turned on the performance drops a bit but other than that we have the same picture as before. The BattleForge engine really likes AMD.  

OVERCLOCKING

As AMD claims that this GPU should overclock well we of course had to try it out. We used the overdrive-feature in the Catalyst-drivers and managed to get a stable overclock at 960 MHz for the GPU (up from 850 MHz). Mark and Tony got their ASUS and Sapphire HD4890’s up to 1 GHz so the claim that these GPU’s overclock well seem well founded.

While we do see an increase in performance it doesn’t sscale exactly as an 12% overclock gives us about 7% increase in Far Cry 2 performance.

POWER USAGE

As you might have noticed from the chart earlier comparing the RV790 with the RV770 the maxiumum board power for the HD4890 has gone up compared to the HD4870.

Power consumption for the complete system excluding the monitor

All thress cards draw approxiumetly the same amount of power.

CONCLUSION

The HD4890 does not blow us away in any regards. It’s a nice update of the HD4870 1 GB without really having any big surprises.The HD4890 aims to plug in a hole between the HD4870 1 GB and the HD4870X2. However, even though the performance is increased, the difference isn’t that big when we look at all the games we have tested today. On average we only see a 5-10% increase in performance with HAWX being a exception. AMD is right to hope that vendors will bring out overclocked versions as this will be a good way to set the HD4890 apart from the HD4870 1 GB.

So what about the GeForce GTX-275? Of course NVIDIA released it today just to steal the HD4890’s thunder. Judging from the benchmarks done by our own Mark it looks like the GTX-275 indeed beats the HD4890 in most benchmarks. However – while the HD4890 will be out today or tomorrow in stores we still will have to wait a while for the GTX-275 to come out in anything else than “limited samples”. We will see how AMD responds to the pricing of the GeForce GTX-275 as we’re sure that they want to make sure the HD4890 stays the “best card” when it comes to performace and pricing.